LLM Council with Deep Research
An AI decision-assistance tool where multiple LLMs evaluate questions from different perspectives, enhanced with Google Deep Research for fact-based analysis. Inspired by Andrej Karpathy's LLM Council.
AILLMsPythonMulti-AgentDeep Research
Overview
The LLM Council is a decision-assistance framework where multiple large language models act as council members, each evaluating a question from their own perspective. After individual analysis, they review each other's responses, vote on outcomes, and a chairman consolidates the final answer.
How It Works
Built on top of Andrej Karpathy's LLM Council concept, this upgraded version adds a Deep Research stage before the council deliberates:
- Research Phase — Each council member submits research questions. A Google Deep Research Agent investigates and produces a comprehensive report.
- Individual Analysis — Each LLM reviews the research and provides their perspective on the question.
- Blind Evaluation — Council members evaluate each other's responses without knowing who wrote what.
- Chairman Summary — The chairman consolidates all perspectives into a final, well-reasoned answer.
Key Features
- Multi-model deliberation using GPT, Claude, Gemini, and other LLMs
- Deep Research integration for fact-grounded responses
- Blind peer evaluation to reduce model bias
- Extensible architecture for adding custom perspectives and new models
What's Next
- Telegram integration with voice note support for on-the-go questions
- Perspective-based prompting (e.g., each councillor evaluates from a different legal specialty)
- Expanded model roster as new LLMs are released